AbsoluteResidual¶

class AbsoluteResidual(Residual):
    r"""Absolute residual conformity scorer using unsigned prediction errors.

    Computes conformity scores as the absolute difference between true
    and predicted values:

    $$s = |y - \hat{y}|$$

    The absolute residuals produce **symmetric** prediction intervals,
    where the lower and upper bounds are equidistant from the point
    prediction.

    Parameters
    ----------
    groups : list of str, dict of str to float, or None, default=None
        Panel group filter (list) or filter with weights (dict). If None,
        all panel groups are included with equal weight.
    components : list of str, dict of str to float, or None, default=None
        Component filter (list) or filter with weights (dict). If None,
        all components are included with equal weight.
    time_weighter : BaseWeighter or None, default=None
        Weighter applied along the time axis (observed timestamps). If None,
        all timestamps contribute equally.
    step_weighter : BaseWeighter or None, default=None
        Weighter applied along the forecasting-step axis. If None, all
        forecasting steps contribute equally.
    vintage_weighter : BaseWeighter or None, default=None
        Weighter applied along the vintage-time axis. If None, all vintages
        contribute equally.

    See Also
    --------
    - [`Residual`][yohou.metrics.conformity.Residual] : Asymmetric variant using signed residuals.
    - [`AbsoluteGammaResidual`][yohou.metrics.conformity.AbsoluteGammaResidual] : Scale-independent symmetric variant.

    Examples
    --------
    >>> import polars as pl
    >>> from datetime import date
    >>> from yohou.metrics.conformity import AbsoluteResidual
    >>> scorer = AbsoluteResidual().fit(
    ...     pl.DataFrame({"time": [date(2020, 1, 1), date(2020, 1, 2)], "y": [1.0, 2.0]})
    ... )
    >>> y_truth = pl.DataFrame({"time": [date(2020, 1, 3), date(2020, 1, 4)], "y": [3.0, 5.0]})
    >>> y_pred = pl.DataFrame({"time": [date(2020, 1, 3), date(2020, 1, 4)], "y": [2.5, 6.0]})
    >>> scores = scorer.score(y_truth, y_pred)
    >>> scores.drop("time").to_series().to_list()
    [0.5, 1.0]

    """

    def __sklearn_tags__(self):
        """Get the tags for this estimator."""
        tags = super().__sklearn_tags__()
        assert tags.scorer_tags is not None
        tags.scorer_tags.symmetric = True
        return tags

    def score(self, y_truth: pl.DataFrame, y_pred: pl.DataFrame, /, **score_params) -> pl.DataFrame:
        """Compute absolute residual conformity scores.

        Parameters
        ----------
        y_truth : pl.DataFrame
            True target values.

        y_pred : pl.DataFrame
            Predicted values.

        Returns
        -------
        pl.DataFrame
            Conformity scores (|y_truth - y_pred|) with "time" column preserved.

        """
        check_is_fitted(self, ["_is_fitted"])

        # Filter out scorer from score_params to avoid conflict with explicit scorer=self
        score_params_filtered = {k: v for k, v in score_params.items() if k != "scorer"}

        # Validate and align (time dropped, returned as context)
        y_truth, y_pred, context = validate_scorer_data(
            self,
            y_truth,
            y_pred,
            **score_params_filtered,
        )

        # Compute scores and reconstruct with time
        scores_values = (y_truth - y_pred).select(pl.all().abs())
        scores = pl.DataFrame({"time": context.time_values}).hstack(scores_values)

        return scores

    def inverse_score(
        self, y_pred: pl.DataFrame, conformity_scores: pl.DataFrame, coverage_rate: float
    ) -> pl.DataFrame:
        """Construct symmetric prediction intervals from absolute conformity scores.

        Parameters
        ----------
        y_pred : pl.DataFrame
            Point predictions, optionally with "time" column.

        conformity_scores : pl.DataFrame
            Absolute conformity scores from calibration set, optionally with "time" column.

        coverage_rate : float
            Desired coverage probability.

        Returns
        -------
        pl.DataFrame
            Symmetric prediction intervals.

        """
        check_is_fitted(self, ["_is_fitted"])

        # Validate and align inputs (time dropped, returned as context for reconstruction)
        y_pred, conformity_scores, context = validate_scorer_data(
            self, y_true=None, y_pred=y_pred, scores=conformity_scores, inverse=True
        )

        # Compute symmetric intervals
        quantile = self._compute_symmetric_quantiles(conformity_scores, coverage_rate)
        lower_bound, upper_bound = y_pred - quantile, y_pred + quantile

        y_pred_interval = self._format_y_pred_interval(lower_bound, upper_bound, coverage_rate)

        # Add time column back
        y_pred_interval = pl.DataFrame({"time": context.time_values}).hstack(y_pred_interval)

        return y_pred_interval